In-Memory Computing
CIM architectures that move computation into the memory array, collapsing the data-movement bottleneck that dominates modern AI accelerators.
He XiaoPh.D. Electrical and Electronic Engineering
Ph.D. Student @ The University of Hong Kong
Bridging advanced AI algorithms with practical high-performance hardware.
I am a Ph.D. student in Electrical and Electronic Engineering at The University of Hong Kong, where I work on making large AI models cheap enough to run everywhere. Before that I was a research scientist intern at PIMIC Inc., working on on-device speech dialogue, 4-bit inference and preliminary ASIC design.
My research sits at the intersection of algorithms and circuits: in-memory computing architectures, ultra-low-bit quantization for large language models, and hardware–software co-optimization for ASIC design. I am passionate about building efficient AI systems that deliver real-world impact on edge devices.
- Affiliation
- The University of Hong Kong
- Advisors
- Prof. Ngai Wong · Prof. Can Li
- Location
- Hong Kong SAR, China
Research
Mission: make large AI models cheap enough to run anywhere.
Ultra-Low-Bit Quantization
Post-training quantization down to 2–4 bits, including trit-plane representations that push LLM inference into genuinely low-precision territory.
Hardware–Software Co-Design
Joint algorithm-and-ASIC design — from kernel fusion and token-aware dataflow to tape-out ready circuits that exploit memristor devices.
News
- 2026.07 Exploring Layer-wise Information Effectiveness for Post-Training Quantization in Small Language Models (LieQ) was published in Findings of ACL 2026.
- 2026.01 A Unified Compute-In-Memory Framework for Multisensory Emotion Recognition was published at ASP-DAC 2026 in Hong Kong.
- 2025.09 PTQTP — post-training quantization to trit-planes for large language models — is now available on arXiv.
- 2025.08 Joined the Institute of Microelectronics, Southern University of Science and Technology as a visiting scholar for a tape-out project collaboration (Aug – Oct 2025).
- 2025 Brain-Inspired Quantized Spiking Neural Networks with Efficient Analog Neurons received the best poster award at MIND 2025.
- 2025 Three Chinese invention patents filed at Southwest University were granted: speech recognition with a lightweight Transformer network, memristor-based text sentiment detection, and brain-inspired vision Transformer image restoration for edge ADAS.
- 2025 Served as a teaching assistant for Math 1853 at the Faculty of Engineering, The University of Hong Kong (Fall 2025).
- 2024.09 Started the Ph.D. programme in Electrical and Electronic Engineering at The University of Hong Kong, advised by Prof. Ngai Wong and Prof. Can Li.
- 2024.09 Started reviewing for IEEE TCAS-I, IEEE TCAS-II, AAAI and NeurIPS.
- 2024.09 Joined PIMIC Inc. as a research scientist intern, working on on-device speech-dialogue LLMs, 4-bit inference and preliminary ASIC design (Sep 2024 – Aug 2025).
- 2024 Brain-Inspired Recognition System Based on Multimodal In-Memory Computing Framework for Edge AI was published in IEEE TCAS-I 71(5).
- 2024 Efficient Low-Bit Neural Network With Memristor-Based Reconfigurable Circuits was published in IEEE TCAS-II 71(1).
- 2023 Memristor-Based Light-Weight Transformer Circuit Implementation for Speech Recognizing was published in IEEE JETCAS 13(1).
- 2023 Pure-Attention-Based Multifunction Memristive Neuromorphic Circuit and System was published in the International Journal of Bifurcation and Chaos 33(9).
Education & Work
-
Sep 2016 – Jun 2020
Southwest University
B.Eng. in Automation
Advisor: Prof. Xiaofang Hu
-
Jul 2019 – Nov 2021
The University of Western Australia
M.Sc. in Electrical and Electronic Engineering
Advisor: Prof. Wen Lei
-
Jun 2022 – Jan 2024
Research Assistant & Team Leader
School of Artificial Intelligence, Southwest University
-
Sep 2024 – Aug 2028
The University of Hong Kong
Ph.D. in Electrical and Electronic Engineering
Advisors: Prof. Ngai Wong, Prof. Can Li
-
Sep 2024 – Aug 2025
Research Scientist Intern – Efficient AI & Edge Deployment
PIMIC Inc. (USA), Hong Kong Science and Technology Park
- Led development of an on-device speech-dialogue chatbot enabling LLM interaction on edge hardware
- Conducted model fine-tuning and quantization to achieve real-time 4-bit inference
- Initiated preliminary ASIC research for specialized AI accelerators
-
Aug 2025 – Oct 2025
Visiting Scholar – Tape-out Project Collaboration
Institute of Microelectronics, Southern University of Science and Technology
-
Fall 2025
Teaching Assistant – Math 1853
The University of Hong Kong, Faculty of Engineering
Awards & Service
Honors 2
- The University of Hong Kong PGS
- Chongqing Outstanding Individual
Service & Leadership 4
- Reviewer — IEEE TCAS-I, IEEE TCAS-II, AAAI, NeurIPS Sep 2024 – Present
- Chairman — Youth Volunteers Association, Southwest University Jun 2018 – Jun 2019
- Main Committee Member — Student Association, Southwest University Jun 2018 – Jun 2019
- Volunteer — Serving left-behind children, the elderly, and children with special needs 100+ hours
Publications
Journal Articles 6
-
High-Performance Method and Architecture for Attention Computation in DNN Inference Journal
-
Neuromorphic Vision Restoration Network for Advanced Driver Assistance System Journal
-
Brain-Inspired Recognition System Based on Multimodal In-Memory Computing Framework for Edge AI Journal
-
Efficient Low-Bit Neural Network With Memristor-Based Reconfigurable Circuits Journal
-
Memristor-Based Light-Weight Transformer Circuit Implementation for Speech Recognizing Journal
-
Pure-Attention-Based Multifunction Memristive Neuromorphic Circuit and System Journal
Conference Papers & Preprints 4
-
Exploring Layer-wise Information Effectiveness for Post-Training Quantization in Small Language Models Conference Findings
-
A Unified Compute-In-Memory Framework for Multisensory Emotion Recognition Conference Poster
-
Brain-Inspired Quantized Spiking Neural Networks with Efficient Analog Neurons Conference Best poster
-
PTQTP: Post-Training Quantization to Trit-Planes for Large Language Models Preprint Under review
Patents 3
-
Speech Recognition System and Method Based on a Lightweight Transformer Network Patent Granted
-
Memristor-Based Text Sentiment Detection System and Method Patent Granted
-
Image Restoration Method Based on a Brain-Inspired Vision Transformer and ADAS Edge Devices Patent Granted